​
Overview

The Extract Data Block uses AI capabilities to extract structured data from unstructured text input. By defining a flexible schema, you can specify which fields to extract, their data types, and provide detailed descriptions to guide the AI model in accurately extracting the desired information.

​
Inputs

​
systemPrompt
string

The system prompt to send to the model. Optional. Used to provide high-level guidance to the AI model.

​
context
string
required

The context to extract the specified schema(s) from. Required.

​
Outputs

​
all
object[]

An array of objects containing all extracted data, structured according to the defined schema.

​
[schemaField]
any[]

For each field defined in the schema, an array containing the extracted values for that field. The type will match the schema field type.

​
duration
number

The time in milliseconds that the block took to execute.

​
Editor Settings

​
AI Model
string
default:"gpt-3.5-turbo"

The AI model to use for data extraction. Available models are dynamically populated based on the LLM provider configuration.

​
Schema
object

The schema defining what data to extract. Specify field names, types, and descriptions to guide the extraction.

​
Parameters

​
Temperature
number
default:0

What sampling temperature to use. Higher values like 0.8 make output more random, while lower values like 0.2 make it more focused and deterministic.

​
Top P
number
default:1

Alternative to temperature sampling. Only tokens comprising the top P probability mass are considered. For example, 0.1 means only tokens in the top 10% probability are considered.

​
Use Top P
boolean
default:false

Whether to use top p sampling instead of temperature sampling.

​
Max Tokens
number
default:8192

The maximum number of tokens to generate in the completion.

​
Stop
string

A sequence where the API will stop generating further tokens.

​
Presence Penalty
number

Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model’s likelihood to talk about new topics.

​
Frequency Penalty
number

Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model’s likelihood to repeat the same line verbatim.

​
Seed
number

If specified, OpenAI will make a best effort to sample deterministically, such that repeated requests with the same seed and parameters should return the same result.

​
Advanced

​
Cache Responses
boolean
default:false

If enabled, requests with the same parameters and messages will be cached for immediate responses without an API call.

​
Use for subgraph partial output
boolean

If enabled, streaming responses from this node will be shown in Subgraph nodes that call this graph.

​
Error on invalid output(s)
boolean

If enabled, the node will error if any of the results returned do not conform to the schema.

​
Error Handling

The block will retry failed attempts up to 3 times with exponential backoff:

  • Minimum retry delay: 500ms
  • Maximum retry delay: 5000ms
  • Retry factor: 2.5x
  • Includes randomization
  • Maximum retry time: 5 minutes

Error messages will be logged for:

  • Invalid schema validation
  • Missing LLM provider configuration
  • API errors and timeouts
  • Aborted requests

Always validate the output of the Extract Data block, especially when working with critical data or making important decisions based on the extracted information.

​
Example: Extracting Contact Information

  1. Add an Extract Data block to your flow.
  2. Connect your input text (e.g., an email or document) to the Context input.
  3. Define a schema in the block settings, for example: 
    {
  "name": {"type": "string", "description": "Full name of the person"},
  "email": {"type": "string", "description": "Email address"},
  "phone": {"type": "string", "description": "Phone number"},
  "company": {"type": "string", "description": "Company name"}
}
  4. Select your desired provider and model.
  5. Run your flow. The block will output an object with the extracted information.

​
See Also